Detection of Chewing Motion Using a Glasses Mounted Accelerometer Towards Monitoring of Food Intake Events in the Elderly

A novel way to detect food intake events using a wearable accelerometer is presented in this paper. The accelerometer is mounted on wearable glasses and used to capture the movements of the head. During meals, a person’s chewing motion is clearly visible in the time domain of the captured accelerometer signal. Features are extracted from this signal and a forward feature selection algorithm is used to determine the optimal set of features. Support Vector Machine and Random Forest classifiers are then used to automatically classify between epochs of chewing and non-chewing. Data was collected from 5 volunteers. The Support Vector Machine approach with linear kernel performs best with a detection accuracy of 73.98% \(\pm\) 3.99.